
from IPython.display import Audio, display, HTML
import matplotlib.pyplot as plt
import numpy as np
import os

%load_ext autoreload
%autoreload 1
%aimport wavenet_project
w = wavenet_project

The autoreload extension is already loaded. To reload it, use:
  %reload_ext autoreload


RATE = 44100

def hide_axes(axes):
    for ax in axes:
        ax.axis('off')

for i, filename in enumerate(os.listdir('pred')):
    fig, axes = plt.subplots(3, dpi=100)
    
    mel_path = os.path.join('wavenet_vocoder/egs/mol/dump/lj/logmelspectrogram/norm/eval/',
                            filename)
    truth_path = os.path.join('wavenet_vocoder/egs/mol/dump/lj/logmelspectrogram/norm/eval/',
                              filename.replace('feats', 'wave'))
    pred_path = os.path.join('pred', filename)
    
    mel = np.load(mel_path)
    truth_wave = np.load(truth_path)
    pred_wave = np.load(pred_path)
    
    axes[0].set_title('MusicNet example #%d mel spectrogram' % (i+1))
    axes[0].imshow(mel, aspect='auto', cmap='coolwarm', interpolation='nearest', origin='lower')
    
    axes[1].set_title('MusicNet example #%d waveform' % (i+1))
    axes[1].plot(truth_wave)
    
    axes[2].set_title('Waveform predicted by WaveNet')
    axes[2].plot(pred_wave)
    
    # hide_axes(axes)
    plt.tight_layout()
    plt.show()
    
    display(HTML('<p>MusicNet example #%d</p>' % (i+1)))
    display(Audio(truth_wave, rate=RATE))
    display(HTML('<p>WaveNet prediction</p>'))
    display(Audio(pred_wave, rate=RATE))

Running WaveNet on training examples¶